Humans and AI working together: Crash Course AI #14

Humans and AI working together: Crash Course AI #14


Striking Sound speed CC AI Take 1 Action INTRO Hey, I’m Jabril and welcome to Crash Course AI! It seems like every time I look at the news,
there’s a new article about how AI and automation is going to take everybody’s jobs! I’m starting to wonder if teaching John
Green Bot those things was even a good idea… but there is a way for AI and humans to work
together, besides competing for the same jobs, resources, and game championships. Human-AI teams can use our strengths to help
each other, and collaborate to fill in each other’s weaknesses. Together, we can make better diagnoses, brainstorm
new inventions, or imagine a future where humans and robots are working side-by-side. AIs and humans have skills that can complement
each other, AI can be good at searching through lots of possibilities and making some intelligent
guesses at which one to pick. Plus, AI systems are consistent, and won’t
make mistakes because they’re tired or hungry, like I sometimes do. On the other hand, humans can be good at insight,
creativity, and understanding the nuances of language and behavior. We’ve learned from living in the world and
interacting with each other, so we’re better than AI systems at interpreting social signals. There are many ways that AI could support
us, but a big one is that AI could amplify our decision-making activities with the right
information. For an example of this kind of Human-AI collaboration,
let’s go to the ThoughtBubble. Humans and computers can play games against
each other, but when they join forces, they basically become a superhero dynamic duo. In chess, this particular kind of Human-AI
team game is called advanced chess, cyborg chess, or centaur chess! In centaur chess, the computer does what it’s
great at: it looks several moves ahead and estimates the most promising next moves. And the human does what they’re great at:
they choose among uncertain possibilities based on experience, intuition, and even what
they know about the opponent. Usually, the computer program is in the driver’s
seat for the early game, which is relatively mindless because there aren’t too many move
choices. But humans step in to guide the strategy in
the middle game when things get more complicated. The first centaur chess tournament was held
in 2007, with the winning team led by Anson Williams. Anson’s team, Intagrand, consists of him
and the programs written by the Computer Science and Math experts Yingheng Chen and Nelson
Hernandez. And the cool thing about this team is that
it’s currently considered to be the best chess player in the world, among humans and AI! Intagrand was able to win 2-0 against chess grandmasters, even though none of the team members are grandmasters. And in 2014, multiple AIs played against multiple
centaur chess teams in competitions. Pure AI won 42 games, while centaur teams
won 53 games! It seems like Human-AI collaboration is working
out for chess, so that means it could be promising in other parts of our lives too. Thanks Thought Bubble. Games provide a great constrained environment
to demonstrate the possibilities of AI and, in this case, human-AI collaboration. Chess victories might not seem that significant,
but similar Human-AI collaboration can be applied to other problems. AI takes on the parts that require memorization,
rote response, and following rules. Humans focus on aspects that require nuance,
social understanding, and intuition. For one, AI could help humans make decisions
when we’re dealing with large amounts of complicated information. When a doctor is trying to make a diagnosis,
they try to use their medical experience and intuition to make sense of their patient’s
symptoms and all the published clinical data. AI could help wade through all that data and
highlight the most probable diagnoses, so the doctor can focus their experience and
intuition on choosing from those which is where they’ll be most helpful. Second, AI could help when humans are trying
to come up with new inventions or designs, like a new engineered structure. An AI could apply predetermined physical constraints,
like, for example, how much something should weigh or how much force it should be able
to withstand. This lets the human experts think about the
most practical designs, and could spark new creative ideas. Third, AI could also support and scale-up
interaction between people. It could save people from doing rote mental
tasks, so that they have more time and energy to help. For example, in customer support, virtual
assistants can help answer easy questions about checking on an order or starting a return…
or at least that’s the goal. If you’ve ever tried these systems, you
know that they can sometimes fail in spectacular ways, leaving you mashing the 0 button on
your phone to try and get a representative on the line. And fourth, robots that are AI-enabled but
guided by humans could give people more strength, endurance, or precision to do certain kinds
of work. Some examples of this may be an exosuit worn
by a construction worker, or a remotely-guided search and rescue robot. These devices still need some AI to apply
the right amount of force or navigate effectively, but all of the real decision-making is done
by humans. There are entire research fields, like Human-Computer
Interaction and Human-Robot Interaction, dedicated to investigating and building new AI, Machine
Learning, and Robotics systems to complement and enhance human capabilities. But it’s not just that humans can become
more effective with AI, AI also needs humans to succeed! This whole series we’ve been talking about
how we can program AI to help them exist and learn, but a lot of human-AI interaction is
more subtle. You could’ve supported an AI without even
recognizing it. First, humans can provide AI with meaning
and labels, because we have so much more experience with living in the real world. For example, if you’ve ever edited Wikipedia,
you’ve contributed to Wikipedia-based algorithms such as WikiBrain. Because Wikipedia puts articles into nesting
structures (like how an elephant is a mammal) and because articles link each other, algorithms
can use this structure to understand the meaning-based connection between topics. In fact, when we interact with digital technology,
whether it’s posting content, giving something a thumbs up, following driving directions
on a phone, or typing a text message, we’re often providing training data to help make
AI systems more effective. Without our data, there wouldn’t be recommended
YouTube videos, predictive text messages, or traffic data for the GPS to use in suggesting
a route. But sometimes providing personal data can
be a double-edged sword which we’ll get to in the Algorithmic Bias episode. Secondly, humans can also try to explain an
AI system’s predictions, outputs, and even possible mistakes to other humans, who aren’t
as familiar with AI. As we mentioned in the episode about artificial neural networks, the reasons for an AI producing particular results can be tough to understand. We can see what data go into the program and
which results fit the data well, but it can be hard to know what the hidden layers are
doing to get those results! For example, an algorithm might recommend
denying a customer’s loan request. So a loan officer needs to be able to look
at the input data and algorithm, and then communicate what factors might’ve led to
denial. Many European countries are now making it
a legal right to receive these kinds of explanations. Third, human experts can also inspect algorithms
for fairness to different kinds of people, rather than producing biased results. Bias is a very complicated topic, so we’ll
dive deeper into the nuances in an upcoming episode and lab. And finally, AI doesn’t understand things
like the potential consequences of its mistakes or the moral implications of its decisions. That’s beyond the scope of its programming. It’s a common Sci-Fi trope that an AI built
to minimize suffering might choose to eliminate all life on Earth, because if there’s no
life, there’s no suffering! That’s why humans may want to moderate and
filter AI actions in the world, so we can make sure they line up with societal values,
morals, and thoughtful intentions. The bottom line is that organic and artificial
brains may be better together, and through Crash Course AI, you could be on the way to
becoming one of those experts that works on the helpfulness and fairness of AI systems. In this episode, we didn’t focus on explaining
one specific algorithm or AI technology. Instead, it’s more about where our world
might be going from the AI revolution that’s happening now… besides just “automation
replacing jobs.” We should recognize what data humans are providing
to algorithms. What would it mean if we could claim some
credit for the ways that our data have allowed algorithms to change lives for the better? Or how do we claim more power in cases where
data are being used in potentially harmful or problematic ways? Second, we should think about if and how our
human jobs could be made easier by working with AI — although it’s bound to be complicated. For example, people had similar concerns with
the spread of personal computing and tools like spreadsheets. And yes, spreadsheets automated many bookkeeping
tasks, which put many people out of work. Even though some types of jobs were destroyed,
new accounting jobs were created that involved human-computer collaboration. Technology took over more of the rote math
calculations, and humans focused on the more nuanced and client-facing aspects of accounting
work. Even though this idea can get overblown in
the mass media, AI and automation has and will take people’s jobs. No question. And we don’t want to downplay the impact
that has on people’s lives. But by understanding how AI works, what it’s
good at, and where it struggles, we can also find opportunities to work more effectively
and to create new types of jobs that involve collaboration. Machines can help us do things that we can’t
do as well (or at all) by ourselves. Human-AI collaboration can help us narrow
down complex decision trees and make better choices. Human-Robot collaboration has the potential
to give us super strength or resilience. Different kinds of AI will impact the world
in powerful ways, but not without costs. So it’s up to us to decide which costs are
worth it, how to minimize harm, and create a future we want to live in. Thanks for watching, I’ll see you next
week. Crash Course AI is produced in association
with PBS Digital Studios. If you want to help keep Crash Course free
for everyone, forever, you can join our community on Patreon. To think more about the complicated lines
between AI and humans, check out this video from Crash Course Philosophy.